Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

HyperCOD: The First Challenging Benchmark and Baseline for Hyperspectral Camouflaged Object Detection

Shuyan Bai, Tingfa Xu, Peifu Liu, Yuhao Qiu, Huiyan Bai, Huan Chen, Yanyan Peng, Jianan Li

TL;DR

This work introduces HyperCOD, the first large-scale hyperspectral camouflaged object detection benchmark, featuring $350$ hyperspectral images with $200$ bands (400–1000 nm) at $1240\times1680$ resolution across diverse scenes, and a Spectral-Spatial SAM framework (HSC-SAM) to bridge hyperspectral data with SAM. The core contributions are the Spectral-Spatial Decomposition Module (SSDM) that decouples hyperspectral input into a CIEXYZ spatial map and a spectral saliency prompt, the Spectral-Guided Token Dropout (SGTD) for efficient background suppression, and the Fusion Detail Enhancer (FDE) for boundary refinement, all integrated within a SAM-based decoder. Extensive experiments show HSC-SAM achieves state-of-the-art results on HyperCOD (e.g., MAE $=0.0017$ and Adp-F $\approx 0.681$) and generalizes well to other HS datasets, while maintaining competitive computational efficiency ($\approx$11.7M parameters, 94.2G FLOPs). HyperCOD and HSC-SAM thus provide a solid foundation to accelerate hyperspectral camouflage detection research and enable robust real-world applications that leverage spectral information for camouflage breakage.

Abstract

RGB-based camouflaged object detection struggles in real-world scenarios where color and texture cues are ambiguous. While hyperspectral image offers a powerful alternative by capturing fine-grained spectral signatures, progress in hyperspectral camouflaged object detection (HCOD) has been critically hampered by the absence of a dedicated, large-scale benchmark. To spur innovation, we introduce HyperCOD, the first challenging benchmark for HCOD. Comprising 350 high-resolution hyperspectral images, It features complex real-world scenarios with minimal objects, intricate shapes, severe occlusions, and dynamic lighting to challenge current models. The advent of foundation models like the Segment Anything Model (SAM) presents a compelling opportunity. To adapt the Segment Anything Model (SAM) for HCOD, we propose HyperSpectral Camouflage-aware SAM (HSC-SAM). HSC-SAM ingeniously reformulates the hyperspectral image by decoupling it into a spatial map fed to SAM's image encoder and a spectral saliency map that serves as an adaptive prompt. This translation effectively bridges the modality gap. Extensive experiments show that HSC-SAM sets a new state-of-the-art on HyperCOD and generalizes robustly to other public HSI datasets. The HyperCOD dataset and our HSC-SAM baseline provide a robust foundation to foster future research in this emerging area.

HyperCOD: The First Challenging Benchmark and Baseline for Hyperspectral Camouflaged Object Detection

TL;DR

This work introduces HyperCOD, the first large-scale hyperspectral camouflaged object detection benchmark, featuring hyperspectral images with bands (400–1000 nm) at resolution across diverse scenes, and a Spectral-Spatial SAM framework (HSC-SAM) to bridge hyperspectral data with SAM. The core contributions are the Spectral-Spatial Decomposition Module (SSDM) that decouples hyperspectral input into a CIEXYZ spatial map and a spectral saliency prompt, the Spectral-Guided Token Dropout (SGTD) for efficient background suppression, and the Fusion Detail Enhancer (FDE) for boundary refinement, all integrated within a SAM-based decoder. Extensive experiments show HSC-SAM achieves state-of-the-art results on HyperCOD (e.g., MAE and Adp-F ) and generalizes well to other HS datasets, while maintaining competitive computational efficiency (11.7M parameters, 94.2G FLOPs). HyperCOD and HSC-SAM thus provide a solid foundation to accelerate hyperspectral camouflage detection research and enable robust real-world applications that leverage spectral information for camouflage breakage.

Abstract

RGB-based camouflaged object detection struggles in real-world scenarios where color and texture cues are ambiguous. While hyperspectral image offers a powerful alternative by capturing fine-grained spectral signatures, progress in hyperspectral camouflaged object detection (HCOD) has been critically hampered by the absence of a dedicated, large-scale benchmark. To spur innovation, we introduce HyperCOD, the first challenging benchmark for HCOD. Comprising 350 high-resolution hyperspectral images, It features complex real-world scenarios with minimal objects, intricate shapes, severe occlusions, and dynamic lighting to challenge current models. The advent of foundation models like the Segment Anything Model (SAM) presents a compelling opportunity. To adapt the Segment Anything Model (SAM) for HCOD, we propose HyperSpectral Camouflage-aware SAM (HSC-SAM). HSC-SAM ingeniously reformulates the hyperspectral image by decoupling it into a spatial map fed to SAM's image encoder and a spectral saliency map that serves as an adaptive prompt. This translation effectively bridges the modality gap. Extensive experiments show that HSC-SAM sets a new state-of-the-art on HyperCOD and generalizes robustly to other public HSI datasets. The HyperCOD dataset and our HSC-SAM baseline provide a robust foundation to foster future research in this emerging area.
Paper Structure (30 sections, 17 equations, 7 figures, 6 tables)

This paper contains 30 sections, 17 equations, 7 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Camouflaged Object Detection Datasets.
  4. Camouflaged Object Detection Methods.
  5. SAM-based Methods.
  6. HyperCOD dataset
  7. Comparison with Existing Datasets.
  8. Data Collection.
  9. Annotation Process.
  10. Challenge Attributes.
  11. Statistical Analysis.
  12. Method
  13. Spectral-Spatial Decomposition Module
  14. Color Space Mapping.
  15. Spectral Saliency Generator.
  16. ...and 15 more sections

Figures (7)

  • Figure 1: While RGB images show near-indistinguishable camouflage, hyperspectral signatures reveal significant differences, highlighting spectral advantages for COD.
  • Figure 2: Representative samples and corresponding ground truth masks from five challenging scenarios.
  • Figure 3: Statistics of the HyperCOD Dataset.
  • Figure 4: Overview of the proposed HSC-SAM framework. Spatial features $\boldsymbol{I}_M$ are used in the SAM image encoder, while spectral saliency $\boldsymbol{I}_S$ guides prompt encoding and token pruning. Joint spatial-spectral learning enhances object localization, and a Fusion Detail Enhancer (FDE) refines boundary details for accurate segmentation.
  • Figure 5: Qualitative results on HyperCOD. HSC-SAM offers clearer contours in challenging camouflage scenarios.
  • ...and 2 more figures